Determine range of validity in parameter space for a trained `EIMInteraction`, given tolerances in observable

[beykyle]

This would allow a user to determine the region of parameter space in which to confidently use an EIM reconstruction if it is smaller than the training range, and to find ways to optimize the size of region of validity.

[beykyle]

I would argue that the use of the new argument expl_var_ratio_cutoff in InteractionEIM.__init__ makes this obsolete (see #64). This argument sets the fraction of variance of the training space that must be explained by the principal components that are kept. The number of reduced basis elements are thus chosen dynamically. I'm going to close this.